Op namespace. More...

Namespaces
	debug
	This namespace includes several methods for debugging parallel communication patterns.

	mpi
	template MPI namespace

	utility
	Utility methods to facilitate common operations.

Classes
class	NLopt
	A op::optimizer implementation for NLopt. More...

struct	NLoptOptions
	Options specific for nlopt. They are made to look like ipopt's interface. More...

class	Go

class	Vector
	Abstracted Optimization Vector container. More...

class	Functional
	Abstracted Objective Functional class. More...

class	Optimizer
	Abstracted Optimizer implementation. More...

class	WaitLoop

Typedefs
using	nlopt_index_type = std::vector< std::size_t >
	Default nlopt type.

using	CallbackFn = std::function< void()>
	Callback function type.

using	ActionFn = std::function< void()>
	Action type we'll use over and over again.

Enumerations
enum	State : int { OTHER, SOLUTION_FOUND = -4, UPDATE_VARIABLES = -3, OBJ_GRAD = -2, OBJ_EVAL = -1 }

Functions
template<typename T >
double	NLoptFunctional (const std::vector< double > &x, std::vector< double > &grad, void *objective_and_optimizer)
	Takes in a op::Functional and computes the objective function and it's gradient as a nlopt function. More...

auto	wrapNLoptFunc (std::function< double(unsigned, const double , double , void *)> func)
	wraps any nltop::function into an objective call and a gradient call More...

template<typename T >
	NLopt (op::Vector< std::vector< double >>, NLoptOptions &, MPI_Comm, utility::CommPattern< T >) -> NLopt< T >

template<class OptType , typename... Args>
std::unique_ptr< OptType >	PluginOptimizer (std::string optimizer_path, Args &&...args)
	Dynamically load an Optimizer. More...

template<typename V , typename T >
auto	ReduceObjectiveFunction (const std::function< V(T)> &local_func, MPI_Op op, MPI_Comm comm=MPI_COMM_WORLD)
	Generate an objective function that performs a global reduction. More...

template<typename T , typename I >
auto	OwnedLocalObjectiveGradientFunction (utility::RankCommunication< T > &info, I &global_ids_to_local, T &reduced_id_list, std::function< std::vector< double >(const std::vector< double > &)> local_obj_grad_func, std::function< double(const std::vector< double > &)> local_reduce_func, MPI_Comm comm=MPI_COMM_WORLD)
	Generate an objective gradient function that takes local variables and reduces them in parallel to locally "owned" variables. More...

template<typename T , typename I , typename ValuesType >
ValuesType	ReturnLocalUpdatedVariables (utility::RankCommunication< T > &info, I &global_ids_to_local, ValuesType &reduced_values)
	Generate update method to propagate owned local variables back to local variables in parallel. More...

template<typename T >
auto	AdvancedRegistration (T &global_ids_on_rank, int root=0, MPI_Comm mpicomm=MPI_COMM_WORLD)

Detailed Description

Op namespace.

Namespace for the OP interface.

Enumeration Type Documentation

enum op::State : int

Define a simple state messaging scheme

n = constraint.size()

-4 => solution found -3 => update variables -2 => obj_grad -1 => obj_eval [0,n-1] => constraint_eval [n,2n-1] => constraint_eval

Definition at line 23 of file op_waitloop.hpp.

Function Documentation

template<typename T >

auto op::AdvancedRegistration	(	T &	global_ids_on_rank,
		int	root = `0`,
		MPI_Comm	mpicomm = `MPI_COMM_WORLD`
	)

AdvancedRegistration procedure given

Returns: CommPattern to use with op::Optimizer

Definition at line 338 of file op.hpp.

template<typename T >

double op::NLoptFunctional	(	const std::vector< double > &	x,
		std::vector< double > &	grad,
		void *	objective_and_optimizer
	)

Takes in a op::Functional and computes the objective function and it's gradient as a nlopt function.

Has the same signature as nlopt::function so we can convert any op::Functional into a nlopt::function

Parameters

[in]	x	the optimization variables (on rank = 0 this is the actual global optimization variables, on other ranks it is the local-view of variables.data())
[in]	grad	the result of the gradient of the function w.r.t. x (on rank 0, this is the global gradient eval, on other ranks it is the owned-local gradient)
[in]	objective	Get FunctionalInfo into this call

for constraints g >= lower_bound, they need to be rewritten as -(g - lower_bound) <= 0

for constraints g <= upper_bound, they need to be rewritten as g - upper_bound < = 0

for constraints g >= lower_bound, they need to be rewritten as -(g - lower_bound) <= 0

for constraints g <= upper_bound, they need to be rewritten as g - upper_bound < = 0

Definition at line 406 of file nlopt_op.hpp.

template<typename T , typename I >

auto op::OwnedLocalObjectiveGradientFunction	(	utility::RankCommunication< T > &	info,
		I &	global_ids_to_local,
		T &	reduced_id_list,
		std::function< std::vector< double >(const std::vector< double > &)>	local_obj_grad_func,
		std::function< double(const std::vector< double > &)>	local_reduce_func,
		MPI_Comm	comm = `MPI_COMM_WORLD`
	)

Generate an objective gradient function that takes local variables and reduces them in parallel to locally "owned" variables.

Parameters

[in]	info	RankCommunication struct for local_variables
[in]	global_ids_to_local	A vector mapping of global ids corresponding to local_variable indices
[in]	local_obj_grad_func	The rank-local gradient contributions corresponding to local_variables
[in]	local_reduce_func	A serial user-defined function computed on "owned" variables over both recieved contributions from other ranks and rank-local gradient contributions
[in]	comm	the MPI communicator

Definition at line 283 of file op.hpp.

template<class OptType , typename... Args>

std::unique_ptr<OptType> op::PluginOptimizer	(	std::string	optimizer_path,
		Args &&...	args
	)

Dynamically load an Optimizer.

Parameters

[in]	optimizer_path	path to dynamically loadable .so plugin
[in]	args	A list of args to pass in for initialization

Definition at line 231 of file op.hpp.

template<typename V , typename T >

auto op::ReduceObjectiveFunction	(	const std::function< V(T)> &	local_func,
		MPI_Op	op,
		MPI_Comm	comm = `MPI_COMM_WORLD`
	)

Generate an objective function that performs a global reduction.

Parameters

[in]	local_func	A user-defined function to compute a rank-local objective-contribution
[in]	op	The MPI reduction operation
[in]	comm	The MPI communicator

Definition at line 257 of file op.hpp.

template<typename T , typename I , typename ValuesType >

ValuesType op::ReturnLocalUpdatedVariables	(	utility::RankCommunication< T > &	info,
		I &	global_ids_to_local,
		ValuesType &	reduced_values
	)

Generate update method to propagate owned local variables back to local variables in parallel.

for the variables a rank owns.. update() should propagate those for the variables an update does not own.. they will be in returned_data returned_remapped_data is a map[local_ids] -> values we want to write it back into our local variable array

Parameters

[in]	info	The RankCommunication information corresponding to local_variable data
[in]	global_ids_to_local	A vector mapping of global ids corresponding to local_variable indices
[in]	reduced_values	The rank-local "owned" variables

Definition at line 313 of file op.hpp.

auto op::wrapNLoptFunc ( std::function< double(unsigned, const double *, double *, void *)> func )

wraps any nltop::function into an objective call and a gradient call

Parameters

[in] func a nlopt::function

Definition at line 40 of file nlopt_op.hpp.

Namespaces

Classes

Typedefs

Enumerations

Functions

Detailed Description

Enumeration Type Documentation

Function Documentation